[LV] Remove assert that VF cannot be scalable in setCostBasedWideningDecision.

Scalarization for scalable vectors is not (yet) supported, so the
LV discards a VF when scalarization is chosen as the widening
decision. It should therefore not assert that the VF is not scalable
when it computes the decision to scalarize.

The code can get here when both the interleave-cost, gather/scatter cost
and scalarization-cost are all illegal. This may e.g. happen for SVE
when the VF=1, to avoid generating `<vscale x 1 x eltty>` types that
the code-generator cannot yet handle.

Reviewed By: david-arm

Differential Revision: https://reviews.llvm.org/D106656

diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index f1d83d9..564812b 100644 (file)
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -7418,8 +7418,6 @@ void LoopVectorizationCostModel::setCostBasedWideningDecision(ElementCount VF) {
         Decision = CM_GatherScatter;
         Cost = GatherScatterCost;
       } else {
-        assert(!VF.isScalable() &&
-               "We cannot yet scalarise for scalable vectors");
         Decision = CM_Scalarize;
         Cost = ScalarizationCost;
       }

diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/sve-gather-scatter.ll b/llvm/test/Transforms/LoopVectorize/AArch64/sve-gather-scatter.ll
index 5738019..7c481d1 100644 (file)
--- a/llvm/test/Transforms/LoopVectorize/AArch64/sve-gather-scatter.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/sve-gather-scatter.ll
@@ -1,4 +1,4 @@
-; RUN: opt -loop-vectorize -dce -instcombine -mtriple aarch64-linux-gnu -mattr=+sve -S %s -scalable-vectorization=on -o - | FileCheck %s
+; RUN: opt -loop-vectorize -dce -instcombine -mtriple aarch64-linux-gnu -mattr=+sve -S %s -scalable-vectorization=preferred -force-target-instruction-cost=1 -o - | FileCheck %s
 
 define void @gather_nxv4i32_ind64(float* noalias nocapture readonly %a, i64* noalias nocapture readonly %b, float* noalias nocapture %c, i64 %n) {
 ; CHECK-LABEL: @gather_nxv4i32_ind64
@@ -122,6 +122,37 @@ for.cond.cleanup:                                 ; preds = %for.inc, %entry
   ret void
 }
 
+
+
+define void @gather_nxv4i32_ind64_stride2(float* noalias nocapture readonly %a, float* noalias nocapture readonly %b, i64 %n) {
+; CHECK-LABEL: @gather_nxv4i32_ind64_stride2
+; CHECK: vector.body:
+; CHECK:      %[[IDX:.*]] = phi i64 [ 0, %vector.ph ], [ %{{.*}}, %vector.body ]
+; CHECK-DAG:  %[[STEP:.*]] = call <vscale x 4 x i64> @llvm.experimental.stepvector.nxv4i64()
+; CHECK-DAG:  %[[IDXSPLATINS:.*]] = insertelement <vscale x 4 x i64> poison, i64 %[[IDX]], i32 0
+; CHECK-DAG:  %[[IDXSPLAT:.*]] = shufflevector <vscale x 4 x i64> %[[IDXSPLATINS]], <vscale x 4 x i64> poison, <vscale x 4 x i32> zeroinitializer
+; CHECK:      %[[ADD:.*]] = add <vscale x 4 x i64> %[[IDXSPLAT]], %[[STEP]]
+; CHECK:      %[[MUL:.*]] = shl <vscale x 4 x i64> %[[ADD]], shufflevector (<vscale x 4 x i64> insertelement (<vscale x 4 x i64> undef, i64 1, i32 0), <vscale x 4 x i64> undef, <vscale x 4 x i32> zeroinitializer)
+; CHECK:      %[[PTRS:.*]] = getelementptr inbounds float, float* %b, <vscale x 4 x i64> %[[MUL]]
+; CHECK:      call <vscale x 4 x float> @llvm.masked.gather.nxv4f32.nxv4p0f32(<vscale x 4 x float*> %[[PTRS]]
+entry:
+  br label %for.body
+
+for.body:                                         ; preds = %entry, %for.body
+  %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
+  %indvars.iv.stride2 = mul i64 %indvars.iv, 2
+  %arrayidx = getelementptr inbounds float, float* %b, i64 %indvars.iv.stride2
+  %0 = load float, float* %arrayidx, align 4
+  %arrayidx2 = getelementptr inbounds float, float* %a, i64 %indvars.iv
+  store float %0, float* %arrayidx2, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond.not = icmp eq i64 %indvars.iv.next, %n
+  br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:                                 ; preds = %for.cond.cleanup.loopexit, %entry
+  ret void
+}
+
 !0 = distinct !{!0, !1, !2, !3, !4, !5}
 !1 = !{!"llvm.loop.mustprogress"}
 !2 = !{!"llvm.loop.vectorize.width", i32 4}